search

Bartzi / stn-ocr

Code for the paper STN-OCR: A single Neural Network for Text Detection and Text Recognition
https://arxiv.org/abs/1707.08831
GNU General Public License v3.0
499 stars 137 forks source link

test on dataset of my own #17

Open whulc opened 6 years ago

whulc commented 6 years ago

I try to evaluate the FSNS using my own datasets. It seems that the image size must be 150*150, and there is something wrong with the .csv file, which I create as your requirement. Can you tell me the requirement of test images and give me an example of the csv file for FSNS?

Bartzi commented 6 years ago

Normally the FSNS input file size is: 600x150 (as there are always 4 views of a sign at once). If you use the command line switch --is-original-fsns, the code will assume this input size. On some experiments we used only one view, this is why the code also supports 150*150 as input size. An example of the csv file format can be found here (bullet point 3)

Nevertheless I recommend that you use our new code, which can be found here. This uses a different library, but still works the same way.